sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 250 insertions, 0 deletions

diff --git a/crypto/async_tx/raid6test.c b/crypto/async_tx/raid6test.c
new file mode 100644
index 000000000..d3fbee1e0
--- /dev/null
+++ b/crypto/async_tx/raid6test.c
@@ -0,0 +1,250 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * asynchronous raid6 recovery self test
+ * Copyright (c) 2009, Intel Corporation.
+ *
+ * based on drivers/md/raid6test/test.c:
+ * 	Copyright 2002-2007 H. Peter Anvin
+ */
+#include <linux/async_tx.h>
+#include <linux/gfp.h>
+#include <linux/mm.h>
+#include <linux/random.h>
+#include <linux/module.h>
+
+#undef pr
+#define pr(fmt, args...) pr_info("raid6test: " fmt, ##args)
+
+#define NDISKS 64 /* Including P and Q */
+
+static struct page *dataptrs[NDISKS];
+unsigned int dataoffs[NDISKS];
+static addr_conv_t addr_conv[NDISKS];
+static struct page *data[NDISKS+3];
+static struct page *spare;
+static struct page *recovi;
+static struct page *recovj;
+
+static void callback(void *param)
+{
+	struct completion *cmp = param;
+
+	complete(cmp);
+}
+
+static void makedata(int disks)
+{
+	int i;
+
+	for (i = 0; i < disks; i++) {
+		get_random_bytes(page_address(data[i]), PAGE_SIZE);
+		dataptrs[i] = data[i];
+		dataoffs[i] = 0;
+	}
+}
+
+static char disk_type(int d, int disks)
+{
+	if (d == disks - 2)
+		return 'P';
+	else if (d == disks - 1)
+		return 'Q';
+	else
+		return 'D';
+}
+
+/* Recover two failed blocks. */
+static void raid6_dual_recov(int disks, size_t bytes, int faila, int failb,
+		struct page **ptrs, unsigned int *offs)
+{
+	struct async_submit_ctl submit;
+	struct completion cmp;
+	struct dma_async_tx_descriptor *tx = NULL;
+	enum sum_check_flags result = ~0;
+
+	if (faila > failb)
+		swap(faila, failb);
+
+	if (failb == disks-1) {
+		if (faila == disks-2) {
+			/* P+Q failure.  Just rebuild the syndrome. */
+			init_async_submit(&submit, 0, NULL, NULL, NULL, addr_conv);
+			tx = async_gen_syndrome(ptrs, offs,
+					disks, bytes, &submit);
+		} else {
+			struct page *blocks[NDISKS];
+			struct page *dest;
+			int count = 0;
+			int i;
+
+			BUG_ON(disks > NDISKS);
+
+			/* data+Q failure.  Reconstruct data from P,
+			 * then rebuild syndrome
+			 */
+			for (i = disks; i-- ; ) {
+				if (i == faila || i == failb)
+					continue;
+				blocks[count++] = ptrs[i];
+			}
+			dest = ptrs[faila];
+			init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL,
+					  NULL, NULL, addr_conv);
+			tx = async_xor(dest, blocks, 0, count, bytes, &submit);
+
+			init_async_submit(&submit, 0, tx, NULL, NULL, addr_conv);
+			tx = async_gen_syndrome(ptrs, offs,
+					disks, bytes, &submit);
+		}
+	} else {
+		if (failb == disks-2) {
+			/* data+P failure. */
+			init_async_submit(&submit, 0, NULL, NULL, NULL, addr_conv);
+			tx = async_raid6_datap_recov(disks, bytes,
+					faila, ptrs, offs, &submit);
+		} else {
+			/* data+data failure. */
+			init_async_submit(&submit, 0, NULL, NULL, NULL, addr_conv);
+			tx = async_raid6_2data_recov(disks, bytes,
+					faila, failb, ptrs, offs, &submit);
+		}
+	}
+	init_completion(&cmp);
+	init_async_submit(&submit, ASYNC_TX_ACK, tx, callback, &cmp, addr_conv);
+	tx = async_syndrome_val(ptrs, offs,
+			disks, bytes, &result, spare, 0, &submit);
+	async_tx_issue_pending(tx);
+
+	if (wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)) == 0)
+		pr("%s: timeout! (faila: %d failb: %d disks: %d)\n",
+		   __func__, faila, failb, disks);
+
+	if (result != 0)
+		pr("%s: validation failure! faila: %d failb: %d sum_check_flags: %x\n",
+		   __func__, faila, failb, result);
+}
+
+static int test_disks(int i, int j, int disks)
+{
+	int erra, errb;
+
+	memset(page_address(recovi), 0xf0, PAGE_SIZE);
+	memset(page_address(recovj), 0xba, PAGE_SIZE);
+
+	dataptrs[i] = recovi;
+	dataptrs[j] = recovj;
+
+	raid6_dual_recov(disks, PAGE_SIZE, i, j, dataptrs, dataoffs);
+
+	erra = memcmp(page_address(data[i]), page_address(recovi), PAGE_SIZE);
+	errb = memcmp(page_address(data[j]), page_address(recovj), PAGE_SIZE);
+
+	pr("%s(%d, %d): faila=%3d(%c)  failb=%3d(%c)  %s\n",
+	   __func__, i, j, i, disk_type(i, disks), j, disk_type(j, disks),
+	   (!erra && !errb) ? "OK" : !erra ? "ERRB" : !errb ? "ERRA" : "ERRAB");
+
+	dataptrs[i] = data[i];
+	dataptrs[j] = data[j];
+
+	return erra || errb;
+}
+
+static int test(int disks, int *tests)
+{
+	struct dma_async_tx_descriptor *tx;
+	struct async_submit_ctl submit;
+	struct completion cmp;
+	int err = 0;
+	int i, j;
+
+	recovi = data[disks];
+	recovj = data[disks+1];
+	spare  = data[disks+2];
+
+	makedata(disks);
+
+	/* Nuke syndromes */
+	memset(page_address(data[disks-2]), 0xee, PAGE_SIZE);
+	memset(page_address(data[disks-1]), 0xee, PAGE_SIZE);
+
+	/* Generate assumed good syndrome */
+	init_completion(&cmp);
+	init_async_submit(&submit, ASYNC_TX_ACK, NULL, callback, &cmp, addr_conv);
+	tx = async_gen_syndrome(dataptrs, dataoffs, disks, PAGE_SIZE, &submit);
+	async_tx_issue_pending(tx);
+
+	if (wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)) == 0) {
+		pr("error: initial gen_syndrome(%d) timed out\n", disks);
+		return 1;
+	}
+
+	pr("testing the %d-disk case...\n", disks);
+	for (i = 0; i < disks-1; i++)
+		for (j = i+1; j < disks; j++) {
+			(*tests)++;
+			err += test_disks(i, j, disks);
+		}
+
+	return err;
+}
+
+
+static int __init raid6_test(void)
+{
+	int err = 0;
+	int tests = 0;
+	int i;
+
+	for (i = 0; i < NDISKS+3; i++) {
+		data[i] = alloc_page(GFP_KERNEL);
+		if (!data[i]) {
+			while (i--)
+				put_page(data[i]);
+			return -ENOMEM;
+		}
+	}
+
+	/* the 4-disk and 5-disk cases are special for the recovery code */
+	if (NDISKS > 4)
+		err += test(4, &tests);
+	if (NDISKS > 5)
+		err += test(5, &tests);
+	/* the 11 and 12 disk cases are special for ioatdma (p-disabled
+	 * q-continuation without extended descriptor)
+	 */
+	if (NDISKS > 12) {
+		err += test(11, &tests);
+		err += test(12, &tests);
+	}
+
+	/* the 24 disk case is special for ioatdma as it is the boundary point
+	 * at which it needs to switch from 8-source ops to 16-source
+	 * ops for continuation (assumes DMA_HAS_PQ_CONTINUE is not set)
+	 */
+	if (NDISKS > 24)
+		err += test(24, &tests);
+
+	err += test(NDISKS, &tests);
+
+	pr("\n");
+	pr("complete (%d tests, %d failure%s)\n",
+	   tests, err, err == 1 ? "" : "s");
+
+	for (i = 0; i < NDISKS+3; i++)
+		put_page(data[i]);
+
+	return 0;
+}
+
+static void __exit raid6_test_exit(void)
+{
+}
+
+/* when compiled-in wait for drivers to load first (assumes dma drivers
+ * are also compiled-in)
+ */
+late_initcall(raid6_test);
+module_exit(raid6_test_exit);
+MODULE_AUTHOR("Dan Williams <dan.j.williams@intel.com>");
+MODULE_DESCRIPTION("asynchronous RAID-6 recovery self tests");
+MODULE_LICENSE("GPL");